10. P2D2: A Mechanism forPrivacy-Preserving Data

Dissemination

Bharat BhargavaDepartment of Computer Sciences

Purdue University

With contributions from Prof. Leszek Lilien and Dr. Yuhui Zhong

Supported in part by NSF grants IIS-0209059 and IIS-0242840.

12/21/05 2

P2D2 - Mechanism for Privacy-Preserving Data Dissemination

Outline1) Introduction

1.1) Interactions and Trust1.2) Building Trust1.3) Trading Weaker Partner’s Privacy Loss for Stronger

Partner’s Trust Gain1.4) Privacy-Trust Tradeoff and Dissemination of

Private Data1.5) Recognition of Need for Privacy Guarantees

2) Problem and Challenges2.1) The Problem2.2) Trust Model2.3) Challenges

3) Proposed Approach: Privacy-Preserving Data Dissemination (P2D2) Mechanism3.1) Self-descriptive Bundles3.2) Apoptosis of Bundles3.3) Context-sensitive Evaporation of Bundles

4) Prototype Implementation5) Conclusions6) Future Work

12/21/05 3

1) Introduction1.1) Interactions and Trust

Trust – new paradigm of security Replaces/enhances CIA (confid./integr./availab.)

Adequate degree of trust required in interactions In social or computer-based interactions:

From a simple transaction to a complex collaboration Must build up trust w.r.t. interaction partners

Human or artificial partners Offline or online

We focus on asymmetric trust relationships:One partner is “weaker,” another is “stronger” Ignoring “same-strength” partners:

Individual to individual, most B2B,

12/21/05 4

1.2) Building Trust (1)

a) Building Trust By Weaker Partners

Means of building trust by weaker partner in his strongeer (often institutional) partner (offline and online):

Ask around Family, friends, co-workers, …

Check partner’s history and stated philosophy Accomplishments, failures and associated recoveries, … Mission, goals, policies (incl. privacy policies), …

Observe partner’s behavior Trustworthy or not, stable or not, … Problem: Needs time for a fair judgment

Check reputation databases Better Business Bureau, consumer advocacy groups, …

Verify partner’s credentials Certificates and awards, memberships in trust-building

organizations (e.g., BBB), … Protect yourself against partner’s misbehavior

Trusted third-party, security deposit, prepayment,, buying insurance, …

12/21/05 5

1.2) Building Trust (2) b) Building Trust by Stronger Partners

Means of building trust by stronger partner in her weaker (often individual) partner (offline and online):

Business asks customer for a payment for goods or services Bank asks for private information Mortgage broker checks applicant’s credit history Authorization subsystem on a computer observes partner’s

behavior Trustworthy or not, stable or not, … Problem: Needs time for a fair judgment

Computerized trading system checks reputation databases e-Bay, PayPal, …

Computer system verifies user’s digital credentials Passwords, magnetic and chip cards, biometrics, …

Business protects itself against customer’s misbehavior Trusted third-party, security deposit, prepayment,, buying

insurance, …

12/21/05 6

1.3) Trading Weaker Partner’s Privacy Loss for Stronger Partner’s Trust Gain

In all examples of Building Trust by Stronger Partners but the first (payments):Weaker partner trades his privacy loss for his trust gain as perceived by stronger partner

Approach to trading privacy for trust:[Zhong and Bhargava, Purdue]

Formalize the privacy-trust tradeoff problem Estimate privacy loss due to disclosing a credential

set Estimate trust gain due to disclosing a credential set Develop algorithms that minimize privacy loss for

required trust gain Bec. nobody likes loosing more privacy than necessary

12/21/05 7

1.4) Privacy-Trust Tradeoff andDissemination of Private Data

Dissemination of private data Related to trading privacy for trust:

Examples above Not related to trading privacy for trust:

Medical records Research data Tax returns …

Private data dissemination can be: Voluntary

When there’s a sufficient competition for services or goods Pseudo-voluntary

Free to decline… and loose service E.g. a monopoly or demand exceeding supply)

Mandatory Required by law, policies, bylaws, rules, etc.

12/21/05 8

Dissemination of Private Datais Critical

Reasons: Fears/threats of privacy violations reduce trust Reduced trust leads to restrictions on interactions

In the extreme:refraining from interactions, even self-imposed isolation

Very high social costs of lost (offline and online) interaction opportunities

Lost business transactions, opportunities Lost research collaborations Lost social interactions …

=> Without privacy guarantees, pervasive computing will never be realized

People will avoid interactions with pervasive devices / systems Fear of opportunistic sensor networks self-organized by electronic

devices around them – can help or harm people in their midst

12/21/05 9

1.5) Recognition of Needfor Privacy Guarantees (1)

By individuals [Ackerman et al. ‘99]

99% unwilling to reveal their SSN 18% unwilling to reveal their… favorite TV show

By businesses Online consumers worrying about revealing

personal dataheld back $15 billion in online revenue in 2001

By Federal government Privacy Act of 1974 for Federal agencies Health Insurance Portability and Accountability Act

of 1996 (HIPAA)

12/21/05 10

1.5) Recognition of Need for Privacy Guarantees (2)

By computer industry research Microsoft Research

The biggest research challenges:According to Dr. Rick Rashid, Senior Vice President for Research

Reliability / Security / Privacy / Business Integrity Broader: application integrity (just “integrity?”)

=> MS Trustworthy Computing Initiative Topics include: DRM—digital rights management (incl.

watermarking surviving photo editing attacks), software rights protection, intellectual property and content protection, database privacy and p.-p. data mining, anonymous e-cash, anti-spyware

IBM (incl. Privacy Research Institute) Topics include: pseudonymity for e-commerce, EPA and EPAL—

enterprise privacy architecture and language, RFID privacy, p.-p. video surveillance, federated identity management (for enterprise federations), p.-p. data mining and p.-p.mining of association rules, Hippocratic (p.-p.) databases, online privacy monitoring

12/21/05 11

1.5) Recognition of Need for Privacy Guarantees (3)

By academic researchers CMU and Privacy Technology Center

Latanya Sweeney (k-anonymity, SOS—Surveillance of Surveillances, genomic privacy)

Mike Reiter (Crowds – anonymity) Purdue University – CS and CERIAS

Elisa Bertino (trust negotiation languages and privacy) Bharat Bhargava (privacy-trust tradeoff, privacy metrics, p.-p. data

dissemination, p.-p. location-based routing and services in networks) Chris Clifton (p.-p. data mining)

UIUC Roy Campbell (Mist – preserving location privacy in pervasive

computing) Marianne Winslett (trust negotiation w/ controled release of private

credentials) U. of North Carolina Charlotte

Xintao Wu, Yongge Wang, Yuliang Zheng (p.-p. database testing and data mining)

12/21/05 12

2) Problem and Challenges2.1) The Problem (1)

“Guardian:”Entity entrusted by private data owners with collection, processing, storage, or transfer of their data

owner can be an institution or a system owner can be a guardian for her own private data

Guardians allowed or required to share/disseminate private data With owner’s explicit consent Without the consent as required by law

For research, by a court order, etc.

“Data”(Private Data)

Guardian 2Second Level

Guardian 1 Original Guardian

Guardian 3

Guardian 5Third-level

Guardian 6Guardian 4

“Owner”(Private Data

Owner)

12/21/05 13

2.1) The Problem (2)

Guardian passes private data to another guardian in a data dissemination chain Chain within a graph (possibly cyclic)

Sometimes owner privacy preferences not transmitted due to neglect or failure Risk grows with chain length and milieu

fallibility and hostility If preferences lost, even honest receiving

guardian unable to honor them

12/21/05 14

2.2) Trust Model Owner builds trust in Primary Guardian (PG)

As shown in Building Trust by Weaker Partners Trusting PG means:

Trusting the integrity of PG data sharing policies and practices Transitive trust in data-sharing partners of PG

PG provides owner with a list of partners for private data dissemination (incl. info which data PG plans to share, with which partner, and why)

OR: PG requests owner’s permission before any private data

dissemination (request must incl. the same info as required for the list)

OR: A hybrid of the above two

E.g., PG provides list for next-level partners AND each second- and lower-level guardian requests owner’s permission before any further private data dissemination

12/21/05 15

2.3) Challenges Ensuring that owner’s metadata are never

decoupled from his data Metadata include owner’s privacy preferences

Efficient protection in a hostile milieu Threats - examples

Uncontrolled data dissemination Intentional or accidental data corruption, substitution,

or disclosure Detection of data or metadata loss Efficient data and metadata recovery

Recovery by retransmission from the original guardian is most trustworthy

12/21/05 16

3) Proposed Approach: Privacy-Preserving Data Dissemination (P2D2) Mechanism

3.1) Design self-descriptive bundles- bundle = private data + metadata- self-descriptive bec. includes metadata

3.2) Construct a mechanism for apoptosis of bundles

- apoptosis = clean self-destruction3.3) Develop context-sensitive

evaporation of bundles

12/21/05 17

Related Work Self-descriptiveness (in diverse contexts)

Meta data model [Bowers and Delcambre, ‘03] KIF — Knowledge Interchange Format [Gensereth and Fikes, ‘92] Context-aware mobile infrastructure [Rakotonirainy, ‘99] Flexible data types [Spreitzer and A. Begel, ‘99]

Use of self-descriptiveness for data privacy Idea mentioned in one sentence[Rezgui, Bouguettaya and Eltoweissy, ‘03]

Term: apoptosis (clean self-destruction) Using apoptosis to end life of a distributed services (esp. in ‘strongly’ active

networks, where each data packet is replaced by a mobile program) [Tschudin, ‘99]

Specification of privacy preferences and policies Platform for Privacy Preferences [Cranor, ‘03] AT&T Privacy Bird [AT&T, ‘04]

12/21/05 18

Bibliography for Related Work

AT&T Privacy Bird Tour: http://privacybird.com/tour/1 2 beta/tour.html. February 2004.

S. Bowers and L. Delcambre. The uni-level description: A uniform framework for representing information in multiple data models. ER 2003-Intl. Conf. on Conceptual Modeling, I.-Y. Song, et al. (Eds.), pp. 45–58, Chicago, Oct. 2003.

L. Cranor. P3P: Making privacy policies more useful. IEEE Security and Privacy, pp. 50–55, Nov./Dec. 2003.

M. Gensereth and R. Fikes. Knowledge Interchange Format. Tech. Rep. Logic-92-1, Stanford Univ., 1992.

A. Rakotonirainy. Trends and future of mobile computing. 10th Intl. Workshop on Database and Expert Systems Applications, Florence, Italy, Sept. 1999.

A. Rezgui, A. Bouguettaya, and M. Eltoweissy. Privacy on the Web: Facts, challenges, and solutions. IEEE Security and Privacy, pp. 40–49, Nov./Dec. 2003.

M. Spreitzer and A. Begel. More flexible data types. Proc. IEEE 8th Workshop on Enabling Technologies (WETICE ’99), pp. 319–324, Stanford, CA, June 1999.

C. Tschudin. Apoptosis - the programmed death of distributed services. In: J. Vitek and C. Jensen, eds., Secure Internet Programming. Springer-Verlag, 1999.

12/21/05 19

3.1) Self-descriptive Bundles Comprehensive metadata

include: owner’s privacy preferences owner’s contact information

guardian’s privacy policies metadata access conditions enforcement specifications data provenance context-dependent and

other components

How to read and write private data

For the original and/or subsequent data guardians

How to verify and modify metadata

How to enforce preferences and policies

Who created, read, modified, or destroyed any portion of data

Application-dependent elementsCustomer trust levels for different contexts

Other metadata elements

Needed to request owner’s access permissions, or notify the owner of any accesses

12/21/05 20

Implementation Issues for Bundles

Provide efficient and effective representation for bundles Use XML – work in progress

Ensure bundle atomicity — metadata can’t be split from data

A simple atomicity solution using asymmetric encryption Destination Guardian (DG) provides public key Source Guardian (or owner) encrypts bundle with public key

Can re-bundle by encrypting different bundle elements with public keys from different DGs

DG applies its corresponding private key to decrypt received bundle Or: decrypts just bundle elements — reveals data DG “needs to know”

Can use digital signature to assure non-repudiation Extra key mgmt effort: requires Source Guardian to provide public key to DG

Deal with insiders making and disseminating illegal copies of data they are authorized to access (but not copy)Considered below (taxonomy)

12/21/05 21

Notification in Bundles (1) Bundles simplify notifying owners or requesting their

consent Contact information in the owner’s contact information Included information

notification = [notif_sender, sender_t-stamp, accessor, access_t-stamp, access_justification, other_info]

request = [req_sender, sender_t-stamp, requestor, requestor_t-stamp, access_justification, other_info]

Notifications / requests sent to owners immediately, periodically, or on demand

Via: automatic pagers / text messaging (SMS) / email messages automatic cellphone calls / stationary phone calls mail

ACK from owner may be required for notifications Messages may be encrypted or digitally signed for security

12/21/05 22

Notification in Bundles (2)

If permission for a request or request_type is: Granted in metadata

=> notify owner Not granted in metadata

=> ask for owner’s permission to access her data For very sensitive data — no default

permissions for requestors are granted Each request needs owner’s permission

12/21/05 23

Transmitting complete bundles between guardians is inefficient They describe all foreseeable aspects of data

privacy For any application and environment

Solution: prune transmitted bundles Adaptively include only needed data and metadata

Maybe, needed “transitively” — for the whole down stream

Use short codes (standards needed) Use application and environment semantics along

the data dissemination chain

Optimization of Bundle Transmission

12/21/05 24

3.2) Apoptosis of Bundles Assuring privacy in data dissemination

Bundle apoptosis vs. private data apoptosisBundle apoptosis is preferable – prevents inferences from metadata

In benevolent settings:use atomic bundles with recovery by retransmission

In malevolent settings:attacked bundle, threatened with disclosure, performs apoptosis

12/21/05 25

Implementation of Apoptosis Implementation

Detectors, triggers and code Detectors – e.g. integrity assertions identifying potential

attacks E.g., recognize critical system and application events

Different kinds of detectors Compare how well different detectors work False positives

Result in superfluous bundle apoptosis Recovery by bundle retransmission Prevent DoS (Denial-of-service) attacks by limiting

repetitions False negatives

May result in disclosure – very high costs (monetary, goodwill loss, etc.)

12/21/05 26

Optimizationof Apoptosis Implementation

Consider alternative detection, trigerring and code implementations

Determine division of labor between detectors, triggers and code

Code must include recovery from false positives Define measures for evaluation of apoptosis

implementations Effectiveness: false positives rate and false negatives rate Costs of false positives (recovery) and false negatives

(disclosures) Efficiency: speed of apoptosis, speed of recovery Robustness (against failures and attacks)

Analyze detectors, triggers and code Select a few candidate implementation techniques for

detectors, triggers and code Evaluation of candidate techniques vis simulate

experiments Prototyping and experimentation in our testbed for

investigating trading privacy for trust

12/21/05 27

3.3) Context-sensitive Evaporation of Bundles

Perfect data dissemination not always desirable Example: Confidential business data shared within

an office but not outside Idea: Context-sensitive bundle evaporation

12/21/05 28

Proximity-based Evaporationof Bundles

Simple case: Bundles evaporate in proportion to their “distance” from their owner Bundle evaporation prevents inferences from

metadata “Closer” guardians trusted more than “distant” ones Illegitimate disclosures more probable at less

trusted “distant” guardians Different distance metrics

Context-dependent

12/21/05 29

Examples of one-dimensional distance metrics Distance ~ business type

Distance ~ distrust level: more trusted entities are “closer”

Multi-dimensional distance metrics Security/reliability as one of dimensions

Examples of Distance Metrics

Insurance Company

B

5

15

5

2

2

1

2

Bank I -Original Guardia

n

Insurance Company

C

Insurance Company A

Bank II

Bank III

Used Car

Dealer 1

Used Car

Dealer 2

Used Car

Dealer 3

If a bank is the original guardian, then:-- any other bank is “closer” than any insurance company-- any insurance company is “closer” than any used car dealer

12/21/05 30

Distorted data reveal less, protects privacy Examples:

accurate data more and more distorted data

Evaporation Implemented asControlled Data Distortion

250 N. Salisbury StreetWest Lafayette, IN

250 N. Salisbury StreetWest Lafayette, IN[home address]

765-123-4567[home phone]

Salisbury StreetWest Lafayette, IN

250 N. University StreetWest Lafayette, IN[office address]

765-987-6543[office phone]

somewhere inWest Lafayette, IN

P.O. Box 1234West Lafayette, IN[P.O. box]

765-987-4321 [office fax]

12/21/05 31

Distorted data reveal less, protects privacy Examples:

accurate data more and more distorted data

Evaporation Implemented asControlled Data Distortion

250 N. Salisbury StreetWest Lafayette, IN

250 N. Salisbury StreetWest Lafayette, IN[home address]

765-123-4567[home phone]

Salisbury StreetWest Lafayette, IN

250 N. University StreetWest Lafayette, IN[office address]

765-987-6543[office phone]

somewhere inWest Lafayette, IN

P.O. Box 1234West Lafayette, IN[P.O. box]

765-987-4321 [office fax]

12/21/05 32

Context-dependent apoptosis for implementing evaporation Apoptosis detectors, triggers, and code enable

context exploitation Conventional apoptosis as a simple case of

data evaporation Evaporation follows a step function

Bundle self-destructs when proximity metric exceeds predefined threshold value

Evaporation asGeneralization of Apoptosis

12/21/05 33

Evaporation could be used for “active” DRM (digital rights management) Bundles with protected contents evaporate when

copied onto ”foreign” media or storage device

Application of Evaporation for DRM

12/21/05 34

4) Prototype Implementation

TERA = Trust-Enhanced Role Assignment(<nr>) – unconditional path[<nr>]– conditional path

(1)

[2a]

(3) User Role

[2b] [2d][2c1]

(2)

(4)

[2c2]

Our experimental system named PRETTY (PRivatE and TrusTed sYstems) Trust mechanisms already implemented

12/21/05 35

Information Flow in PRETTY1) User application sends query to server application.2) Server application sends user information to TERA server for trust

evaluation and role assignment.a) If a higher trust level is required for query, TERA server sends the request

for more user’s credentials to privacy negotiator.b) Based on server’s privacy policies and the credential requirements, privacy

negotiator interacts with user’s privacy negotiator to build a higher level of trust.

c) Trust gain and privacy loss evaluator selects credentials that will increase trust to the required level with the least privacy loss. Calculation considers credential requirements and credentials disclosed in previous interactions.

d) According to privacy policies and calculated privacy loss, user’s privacy negotiator decides whether or not to supply credentials to the server.

3) Once trust level meets the minimum requirements, appropriate roles are assigned to user for execution of his query.

4) Based on query results, user’s trust level and privacy polices, data disseminator determines: (i) whether to distort data and if so to what degree, and (ii) what privacy enforcement metadata should be associated with it.

12/21/05 36

5) Conclusions Intellectual merit

A mechanism for preserving privacy in data dissemination (bundling, apoptosis, evaporation)

Broader impact Educational and research impact: student projects,

faculty collaborations Practical (social, economic, legal, etc.) impact:

Enabling more collaborations Enabling “more pervasive” computing

By reducing fears of privacy invasions Showing new venues for privacy research

Applications Collaboration in medical practice, business, research,

military… Location-based services

Future impact: Potential for extensions enabling “pervasive computing”

Must adapt to privacy preservation, e.g., in opportunistic sensor networks (self-organize to help/harm)

12/21/05 37

6) Future Work Provide efficient and effective representation for

bundles (XML for metadata?) Run experiments on the PRETTY system

Build a complete prototype of proposed mechanism for private data dissemination

Implement Examine implementation impacts:

Measures: Cost, efficiency, trustworthiness, other Optimize bundling, apoptosis and evaporation techniques

Focus on selected application areas Sensor networks for infrastructure monitoring (NSF IGERT

proposal) Healthcare enginering (work for RCHE - Regenstrief

Center for Healthcare Engineering at Purdue)

12/21/05 38

Future Work - Extensions Adopting proposed mechanism for DRM, IRM

(intellectual rights managenment) and proprietary/confidential data

Privacy: Private data – owned by an individual

Intellectual property, trade/diplomatic/military secrets: Proprietary/confidential data – owned by an organization

Custimizing proposed mechanismm for selected pervasive environments, including:

Wireless / Mobile / Sensor networks Incl. opportunistic sens. networks

Impact of proposed mechanism on data quality